命失，一言可正。
$ python app.py
ModuleNotFoundError: No module named 'uvicorn'
$ ai-fix
✗ Failed: No module named 'uvicorn'
Fix (high confidence): uvicorn is not installed in this environment.
→ pip install uvicorn
Apply fix? [Y/n]: y
Running: pip install uvicorn
✓ Done.
Re-running: python app.py
✓ Fixed! Command succeeded.
安顿
pip install ai-fix
请设汝之API密钥：
export ANTHROPIC_API_KEY=sk-ant-... # Claude (recommended — fast + cheap) # or export OPENAI_API_KEY=sk-... # OpenAI fallback
事毕。如是而已。
用之
# Fix the last command that failed ai-fix # Run a command and fix it if it fails ai-fix python app.py ai-fix npm run build ai-fix cargo build ai-fix docker compose up # Skip the confirmation prompt ai-fix -y python app.py # See what it would do without applying ai-fix --dry-run python app.py # Show full error output ai-fix -v npm run build
其运作之理何在？
- 摄之复行尔败令，并录其标准输出与标准错误。
- 送之— 运错误于系统之境，达于克劳德（或GPT-4o-mini）
- 得回 — 有序之修命令，附信度评分
- 施 — 运修之，复行尔初令以验
智见：尔令，全错，尔之OS，Python/Node之版本，及存之项目文件（package.json，pyproject.toml）Cargo.toml等。其返之指令，恰如其分，无冗文，无赘解，需解之文无有。
例
Python之模块阙如
$ python server.py
ModuleNotFoundError: No module named 'fastapi'
$ ai-fix
Fix (high confidence): fastapi is not installed.
→ pip install fastapi
npm构建之败
$ npm run build
sh: tsc: command not found
$ ai-fix
Fix (high confidence): TypeScript compiler not installed.
→ npm install --save-dev typescript
→ npx tsc
权限所阻
$ ./deploy.sh
bash: permission denied: ./deploy.sh
$ ai-fix
Fix (high confidence): Script is not executable.
→ chmod +x deploy.sh
端口已占
$ python manage.py runserver
Error: That port is already in use.
$ ai-fix
Fix (medium confidence): Port 8000 is occupied by another process.
→ lsof -ti:8000 | xargs kill -9
Git推送之拒
$ git push
! [rejected] main -> main (non-fast-forward)
$ ai-fix
Fix (high confidence): Remote has commits not in your local branch.
→ git pull --rebase origin main
→ git push
Docker未行
$ docker ps
Cannot connect to the Docker daemon. Is docker running?
$ ai-fix
Fix (high confidence): Docker daemon is not running.
→ open -a Docker
何不径以谷歌之？
尔已知如何用谷歌。然当尔深陷于构建之中，切换至浏览器，阅三Stack Overflow之答，复归，则费时且断流。
ai-fix 使尔恒处于终端。一言。毕。
自信程度
| 色 | 义 |
|---|---|
| 🟢高 | AI确信——常见之误，明其解。 |
| 🟡 中等 | 或可修复，然视汝之配置而定 |
| 🔴 低 | AI臆测——施用前请审阅 |
壳之支持
自：
- zsh (
~/.zsh_history) - bash (
~/.bash_history) - fish (
~/.local/share/fish/fish_history）
耗
ai-fix默认使用克劳德俳句——最快且最廉之模型。每修正之费，约计三十万分之一不足一厘之一。欲耗一元，须运之三千遍。
隱私
- 唯送：其败命，其误出（截至三千字），尔之OS，言版，及一列项目文名（非内容）
- 未尝传文件之内容、环境之变量、API之钥、路径之外于当前目录之名
要求
- Python 3.11以上
ANTHROPIC_API_KEY或OPENAI_API_KEY
许可
MIT